ABSTRACT
Much research in the medical and health sciences focuses on the distribution of time to the occurrence of some event that represents failure. For example, failure can be the first onset of a particular disease, the recurrence of disease after treatment, or the death of an individual. This chapter describes how mixture models via the expectation-maximisation algorithm can be adopted for the survival analysis of censored failure time data. It considers the use of survival mixture models to analyse mortality data. The survival experience of patients after diagnosis or certain treatments may be described by sub-populations of patients with different risks of mortality. The chapter describes an expectation-conditional maximization-based semi-parametric mixture model approach that adopts a nonparametric specification for the baseline hazard functions in order to relax the parametric constraints. Mixture models have been used to model survival data in a variety of situations.
